Customers are from Mars, Managers are from Venus: Deriving Customer Satisfaction Drivers from Online Reviews
The Internet is host to many sites that collect vast amounts of opinions about products and services. These opinions are expressed in written language, and automatic analysis of the written opinions is known as sentiment analysis or opinion mining. In this paper, the written opinions constitute unstructured input data, which we first transform into semi-structured data using an automated framework for aspect-level sentiment analysis. Second, we model the overall customer satisfaction using a Bayesian approach based on the individual aspect rating of each review. Our probabilistic method enables us to discover the relative importance of each aspect for each individual product or service. Empirical experiments on a data set of online reviews of California State Parks, obtained from tripadvisor.com, show the effectiveness of the proposed framework as applied to the aspect-level sentiment analysis and modeling of customer satisfaction with an accuracy of 88.3% in terms of finding the significant aspects.
PAPER: 16